基于倒谱修正模型的语音基音检测算法  

Speech Pitch Detection Algorithm Based on Modified Cepstrum Model

作　　者：曾毓敏[1] 吴镇扬[1] 

机构地区：[1]东南大学信息科学与工程学院

出　　处：《南京理工大学学报》2007年第4期503-508,共6页Journal of Nanjing University of Science and Technology

基　　金：国家973计划项目(2002CB312102);江苏省高校自然科学基础研究项目(07KJD510110)

摘　　要：该文提出了一种基于修正倒谱模型的改进的倒谱基音检测算法。该算法首先对分帧语音进行10阶线性预测编码(LPC)分析和逆滤波,获得LPC预测残差;然后对残差信号进行倒谱分析,倒谱分析中采用了离散傅里叶变换频谱的高频分量置零的计算措施;最后根据倒谱的特征求得浊音语音的基音周期。仿真检测结果表明:该算法无论对纯净语音,还是对不同加噪情况下的含噪语音,其基音检测结果都明显优于传统倒谱基音检测算法,并且也明显优于基于平均幅度差函数的基音检测算法,而略优于基于自相关函数的基音检测算法。An improved speech pitch detection algorithm based on modified cepstrum model is pro- posed. In the proposed algorithm, a tenorder LPC （linear predictive coding） analysis is performed on a segmented speech, and the segmented speech is filtered by the inverse filter to give the LPC predictive residual. The cepstrum of the predictive residual is calculated with the simple method of the high frequency spectral components of DFT being set to zero. The pitch period of the voiced speech is extracted from the cepstrum of predictive residual. The simulated pitch detection results show that the pitch extraction error of the proposed algorithm is significantly lower than that of the conventional cep- strum based algorithm both for clean speech and different noisy speech. The performance of the proposed algorithm is also much better than that of the average magnitude difference function based pitch detection algorithm and slightly better than that of the basic autocorrelation function based algorithm.

关 键 词：基音 倒谱 线性预测编码 预测残差 

分 类 号：TN912.3[电子电信—通信与信息系统]